Joint Geodesic Upsampling of Depth Images

Abstract

We propose an algorithm utilizing geodesic distances to upsample a low resolution depth image using a registered high resolution color image. Specifically, it computes depth for each pixel in the high resolution image using geodesic paths to the pixels whose depths are known from the low resolution one. Though this is closely related to the all-pairshortest-path problem which has O(n 2 log n) complexity, we develop a novel approximation algorithm whose complexity grows linearly with the image size and achieve realtime performance. We compare our algorithm with the state of the art on the benchmark dataset and show that our approach provides more accurate depth upsampling with fewer artifacts. In addition, we show that the proposed algorithm is well suited for upsampling depth images using binary edge maps, an important sensor fusion application.

An MRF-Based Depth Upsampling: Upsample the Depth Map With Its Own Property

“…We compare our method with the MRF in [1] and its extension non-local means MRF in [10], the image guided anisotropic total generalized variation upsampling in [2], the joint geodesic upsampling in [7], the guided filter in [4], the JBU in [6] and the method in [20] that is denoted as the layer joint bilateral upsampling. The upsampling results are evaluated in root mean square error (RMSE) between the original depth map and the upsampling result.…”

“…Visual comparison on the real data Books from [2]. The first row are (a) the measured groundtruth and results by (b) bicubic interpolation, (c) the crossbased local multipoint filter in [8], (d) the joint geodesic upsampling in [7], (e) the joint bilateral filter in [6], (f) the MRF in [1], (g) the non-local means MRF in [10], (h) the image guided total generalized variation upsampling in [2] and (i) our method. The intensity image together with the input depth map (in the red box) and corresponding error maps are shown in the second row.…”

Lossless embedding of depth hints in JPEG compressed color images

“…Specifically, since the color and depth images are associated with the same scene, edges of the two images are matched at object boundaries [23]. In [8], it is also demonstrated that color image segmentation can be effectively used in reconstructing the depth image from sparse depth pixels.…”

Depth Superresolution by Transduction

“…The second approach exploits a HR intensity image as a depth cue. It assumes that there exist co-occurrence statistics between depth and intensity discontinuities [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23]. The additional intensity image helps align depth boundaries to intensity edges, but this may cause the textures in the intensity image to be transferred to the depth image.…”